Cluster-based control of self-excited thermoacoustic oscillations

We present the first application of cluster-based control (CBC), a data-driven feedback control strategy, to suppress self-excited thermoacoustic oscillations. CBC embeds a single scalar time-series measurement in a low-dimensional feature space, partitions that space into a finite set of clusters, and assigns an actuation amplitude to each cluster. The amplitudes are then optimized with a Nelder-Mead simplex search that minimizes a cost function balancing oscillation suppression against actuation effort. Implemented on both a low-order thermoacoustic model and an experimental Rijke tube, CBC is found to reduce the pressure amplitude by nearly 98% while requiring an order of magnitude less actuation power than conventional open-loop time-periodic forcing. The CBC algorithm converges after only several optimization iterations, cutting the total training and tuning time by more than a factor of five relative to recent machine-learning-based strategies. These results demonstrate that CBC can provide a rapid sample-efficient route to model-free feedback control of self-excited thermoacoustic systems and, more broadly, of nonlinear self-excited oscillators governed by coupled multi-scale interactions.